New approach to specific Alzheimer's disease diagnosis based on plasma biomarkers in a cognitive disorder cohort.

Background: The validation of a combination of plasma biomarkers and demographic variables is required to establish reliable cut-offs for Alzheimer's disease diagnosis (AD).

Methods: Plasma biomarkers (Aβ42/Aβ40, p-Tau181, t-Tau, NfL, GFAP), ApoE genotype, and demographic variables were obtained from a retrospective clinical cohort of cognitive disorders (n = 478). These patients were diagnosed as AD (n = 254) or non-AD (n = 224) according to cerebrospinal fluid (CSF) Aβ42/Aβ40 levels. An analysis using a Ridge logistic regression model was performed to predict the occurrence of AD. The predictive performance of the model was assessed using the observations from a training set (70% of the sample) and validated using a test set (30% of the sample) in each group. Optimum cutoffs for the model were evaluated.

Results: The model including plasma Aβ42/Aβ40, p-Tau181, GFAP, ApoE genotype and age was optimal for predicting CSF Aβ42/Aβ40 positivity (AUC .91, sensitivity .86, specificity .82). The model including only plasma biomarkers (Aβ42/Aβ40, p-Tau181, GFAP) provided reliable results (AUC .88, sensitivity .83, specificity .78). Also, GFAP, individually, showed the best performance in discriminating between AD and non-AD groups (AUC .859). The established cut-offs in a three-range strategy performed satisfactorily for the validated predictive model (probability) and individual plasma GFAP (concentration).

Conclusions: The plasma GFAP levels and the validated predictive model based on plasma biomarkers represent a relevant step toward the development of a potential clinical approach for AD diagnosis, which should be assessed in further research.